ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

gsdx-ogl: implement a fast accurate colclip algo 

The idea is to use a floating texture to accumulate the data and
then do a final postprocessing pass to apply the modulo

v2:
* use bounding box to
* fix vertex corruption issue
* use negative number in shader which allow to use half float (+12
  fps@4x)
This commit is contained in:
Gregory Hainaut 2015-07-24 22:06:35 +02:00
parent aa8f5848d1
commit e026f1bac6
6 changed files with 59 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
plugins/GSdx/GSDeviceOGL.cpp
View File

@ -889,7 +889,7 @@ GSTexture* GSDeviceOGL::CopyOffscreen(GSTexture* src, const GSVector4& sRect, in
}
// Copy a sub part of texture (same as below but force a conversion)
void GSDeviceOGL::CopyRectConv(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r)
void GSDeviceOGL::CopyRectConv(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r, bool at_origin)
{
const GLuint& sid = sTex->GetID();
const GLuint& did = dTex->GetID();
@ -899,7 +899,10 @@ void GSDeviceOGL::CopyRectConv(GSTexture* sTex, GSTexture* dTex, const GSVector4
gl_BindFramebuffer(GL_READ_FRAMEBUFFER, m_fbo_read);
gl_FramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, sid, 0);
if (at_origin)
gl_CopyTextureSubImage2D(did, GL_TEX_LEVEL_0, 0, 0, r.x, r.y, r.width(), r.height());
else
gl_CopyTextureSubImage2D(did, GL_TEX_LEVEL_0, r.x, r.y, r.x, r.y, r.width(), r.height());
gl_BindFramebuffer(GL_READ_FRAMEBUFFER, 0);
@ -924,7 +927,7 @@ void GSDeviceOGL::CopyRect(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r
r.width(), r.height(), 1);
} else {
// Slower copy (conversion is done)
CopyRectConv(sTex, dTex, r);
CopyRectConv(sTex, dTex, r, true);
}
GL_POP();

2
plugins/GSdx/GSDeviceOGL.h
View File

@ -629,7 +629,7 @@ class GSDeviceOGL : public GSDevice
GSTexture* CopyOffscreen(GSTexture* src, const GSVector4& sRect, int w, int h, int format = 0, int ps_shader = 0);
void CopyRect(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r);
void CopyRectConv(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r);
void CopyRectConv(GSTexture* sTex, GSTexture* dTex, const GSVector4i& r, bool at_origin);
void StretchRect(GSTexture* sTex, const GSVector4& sRect, GSTexture* dTex, const GSVector4& dRect, int shader = 0, bool linear = true);
void StretchRect(GSTexture* sTex, const GSVector4& sRect, GSTexture* dTex, const GSVector4& dRect, GLuint ps, bool linear = true);
void StretchRect(GSTexture* sTex, const GSVector4& sRect, GSTexture* dTex, const GSVector4& dRect, GLuint ps, GSBlendStateOGL* bs, bool linear = true);

55
plugins/GSdx/GSRendererOGL.cpp
View File

@ -344,6 +344,14 @@ GSRendererOGL::PRIM_OVERLAP GSRendererOGL::PrimitiveOverlap()
return PRIM_OVERLAP_NO;
}
GSVector4i GSRendererOGL::ComputeBoundingBox(const GSVector2& rtscale, const GSVector2i& rtsize)
{
GSVector4 scale = GSVector4(rtscale.x, rtscale.y);
GSVector4 offset = GSVector4(-1.0f, 1.0f); // Round value
GSVector4 box = m_vt.m_min.p.xyxy(m_vt.m_max.p) + offset.xxyy();
return GSVector4i(box * scale.xyxy()).rintersect(GSVector4i(0, 0, rtsize.x, rtsize.y));
}
void GSRendererOGL::SendDraw(bool require_barrier)
{
GSDeviceOGL* dev = (GSDeviceOGL*)m_dev;
@ -388,6 +396,8 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
GSDrawingEnvironment& env = m_env;
GSDrawingContext* context = m_context;
GSTexture* hdr_rt = NULL;
const GSVector2i& rtsize = ds->GetSize();
const GSVector2& rtscale = ds->GetScale();
@ -497,14 +507,11 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
ps_sel.blend_accu = m_sw_blending && ALPHA.A == 0 && ALPHA.B == 2 && ALPHA.C != 1 && ALPHA.D == 1;
om_bsel.accu = ps_sel.blend_accu;
bool sw_blending_base = m_sw_blending && (free_blend || impossible_blend /*|| ps_sel.blend_accu*/);
bool sw_blending_base = m_sw_blending && (free_blend || impossible_blend || ps_sel.blend_accu);
// Color clip
bool acc_colclip_wrap = false;
if (env.COLCLAMP.CLAMP == 0) {
// Not supported yet in colclip
om_bsel.accu = ps_sel.blend_accu = 0;
acc_colclip_wrap = (m_sw_blending >= ACC_BLEND_CCLIP || sw_blending_base);
if (acc_colclip_wrap) {
ps_sel.colclip = 3;
@ -514,8 +521,6 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
ps_sel.colclip = 1;
GL_INS("COLCLIP ENABLED (blending is %d/%d/%d/%d)", ALPHA.A, ALPHA.B, ALPHA.C, ALPHA.D);
}
} else {
sw_blending_base |= m_sw_blending && ps_sel.blend_accu;
}
bool all_blend_sw;
@ -583,15 +588,11 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
gl_TextureBarrier();
dev->PSSetShaderResource(3, rt);
} else if (DATE) {
// TODO: do I need to clamp the value (if yes how? rintersect with rt?)
GSVector4 si = GSVector4(rtscale.x, rtscale.y);
GSVector4 off = GSVector4(-1.0f, 1.0f); // Round value
GSVector4 b = m_vt.m_min.p.xyxy(m_vt.m_max.p) + off.xxyy();
GSVector4i ri = GSVector4i(b * si.xyxy());
GSVector4i dRect = ComputeBoundingBox(rtscale, rtsize);
// Reduce the quantity of clean function
glScissor( ri.x, ri.y, ri.width(), ri.height() );
GLState::scissor = ri;
glScissor( dRect.x, dRect.y, dRect.width(), dRect.height() );
GLState::scissor = dRect;
// Must be done here to avoid any GL state pertubation (clear function...)
// Create an r32ui image that will containt primitive ID
@ -599,10 +600,8 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
dev->InitPrimDateTexture(rt);
dev->PSSetShaderResource(3, rt);
} else {
GSVector4 s = GSVector4(rtscale.x / rtsize.x, rtscale.y / rtsize.y);
GSVector4 src = (b * s.xyxy()).sat(off.zzyy());
GSVector4 dst = src * 2.0f + off.xxxx();
GSVector4 src = GSVector4(dRect) / GSVector4(rtsize.x, rtsize.y).xyxy();
GSVector4 dst = src * 2.0f - 1.0f;
GSVertexPT1 vertices[] =
{
@ -913,7 +912,15 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
GL_POP();
}
if (env.COLCLAMP.CLAMP == 0 && om_bsel.accu) {
hdr_rt = dev->CreateTexture(rtsize.x, rtsize.y, GL_RGBA16F);
dev->CopyRectConv(rt, hdr_rt, ComputeBoundingBox(rtscale, rtsize), false);
dev->OMSetRenderTargets(hdr_rt, ds, &scissor);
} else {
dev->OMSetRenderTargets(rt, ds, &scissor);
}
if (context->TEST.DoFirstPass())
{
@ -997,10 +1004,22 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
}
}
}
if (DATE_GL42)
if (DATE_GL42) {
dev->RecycleDateTexture();
}
dev->EndScene();
// Warning: EndScene must be called before StretchRect otherwise
// vertices will be overwritten. Trust me you don't want to do that.
if (hdr_rt) {
GSVector4 dRect(ComputeBoundingBox(rtscale, rtsize));
GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
dev->StretchRect(hdr_rt, sRect, rt, dRect, 4, false);
dev->Recycle(hdr_rt);
}
GL_POP();
}

2
plugins/GSdx/GSRendererOGL.h
View File

@ -54,6 +54,8 @@ class GSRendererOGL : public GSRendererHW
PRIM_OVERLAP m_prim_overlap;
GSVector4i ComputeBoundingBox(const GSVector2& rtscale, const GSVector2i& rtsize);
protected:
void EmulateGS();
void SetupIA();

6
plugins/GSdx/res/glsl/tfx_fs.glsl
View File

@ -612,6 +612,12 @@ void ps_main()
ps_fbmask(C);
#if PS_BLEND_ACCU && PS_COLCLIP
// Use negative value to avoid overflow of the texture
if (any(greaterThan(C.rgb, vec3(128.0f)))) {
C.rgb = (C.rgb - 256.0f);
}
#endif
SV_Target0 = C / 255.0f;
SV_Target1 = vec4(alpha_blend);
}

6
plugins/GSdx/res/glsl_source.h
View File

@ -1471,6 +1471,12 @@ static const char* tfx_fs_all_glsl =
"\n"
" ps_fbmask(C);\n"
"\n"
"#if PS_BLEND_ACCU && PS_COLCLIP\n"
" // Use negative value to avoid overflow of the texture\n"
" if (any(greaterThan(C.rgb, vec3(128.0f)))) {\n"
" C.rgb = (C.rgb - 256.0f);\n"
" }\n"
"#endif\n"
" SV_Target0 = C / 255.0f;\n"
" SV_Target1 = vec4(alpha_blend);\n"
"}\n"